Rackspace vs. Amazon — the Cloud Wars

fdicostanzo writes "The folks at Mixpanel are leaving Rackspace's server cloud for Amazon and have left a little note about their reasons. There's been some talk that Rackspace's offering has not been up to snuff once you scale. Analysis suggests that Rackspace's offering still has some advantages however."

Comments

Sometimes you need real hardware

[jwthompson2]

They missed the fact that RackSpace offers hybrid cloud options that Amazon just can't match at this point. Got IO issues? So did GitHub when they were running on Amazon's infrastructure. Know how they solved it? They moved to Rackspace and married the cloud for front-end with physical hardware for their IO intense workloads. It seems to me these guys may just be naive. They've probably only sidestepped their problems for now.

Re:Sometimes you need real hardware

[Chrisq]

They missed the fact that RackSpace offers hybrid cloud options that Amazon just can't match at this point. Got IO issues? So did GitHub when they were running on Amazon's infrastructure. Know how they solved it? They moved to Rackspace and married the cloud for front-end with physical hardware for their IO intense workloads. It seems to me these guys may just be naive. They've probably only sidestepped their problems for now.

To be fair if they have probably solved their problems, in that Amazon cloud is extremely horizontally scalable. It is a typical "throw money at it" solution, like someone who has sent a package by motorcycle courtier solving the problem of shifting 10,000 packages between warehouses by hiring 10,000 motorcycle couriers - but it will probably work for them.

Re:

[mark72005]

Interesting discussion. Perhaps more companies could make business out of their spare resources as Amazon does.

Also funny, in the comments section with GoGrid.com trolling with a $100 coupon code. Way to sweeten the pot...

Colocation?

[Oceanplexian]

The problem with these solutions is they sell you services like a prepaid phone company to abstract the real cost.

My company has done the math and unless you only need the capacity say, 3 hours out of the day, EC2 (and Rackspace) simply can't compete with running your own hardware. We've heard the arguments about hiring engineers, buying servers, and renting space, but even after those expenses you still come out ahead if you have roughly more than 20 machines.

Also, Rackspace and Amazon sell Xen virtualization hosting. The software is open source and freely available if you want to use it for yourself. I just guess "Cloud Hosting" sounds better but it's not that hard to roll a similar setup if you want the scalability.

Re:Colocation?

[codepunk]

Or you need to scale from 100 to 100,000 users in two days.

Re:Colocation?

[fuzzyfuzzyfungus]

I don't think that, with the fairly classical "cloud" guys like Amazon, who offer essentially Virtual Private Server products; but with automated provisioning, the argument has ever been about much more than flexibility. The more abstracted "cloud" people(gmail, google docs, etc.) can make the management complexity argument, since they offer highly abstract services with all the gory details hidden; but Amazon basically just tosses you a Linux VM and leaves you to deal with it from there. The only attraction is that, unlike classic VPS, you don't have to talk to a sales rep, just an API.

Given that with Amazon, all their EC2 stuff is either uber-commodity linux VMs or available for local hosting via the Eucalyptus project(their storage mechanism, etc.) it is possible to adopt a hybrid "base load/burst load" strategy, similar to how the electrical utilities do it. If your operation has a more or less steady base load, you run it on cheap boxes of your own. If you have a load spike, you use the expensive; but quick to spin up, EC2 instances. Since modern virtualization overhead is low, and virtualization is extremely convenient anyway, you don't lose to much, you don't pay Amazon a flexibility fee for things you don't need to be flexible; but you can swiftly pay for additional capacity that works just like your local capacity, and then stop paying when you no longer need it.

If you need long-term, stable levels of service, you'd be insane to buy it from a burst-service company, just as very heavy cell users would be nuts to buy a contractless per-minute plan. Either do it in house or hire a hosting company to do it for you, on a stable basis.

Re:Colocation?

[Anonymous Coward]

We've heard the arguments about hiring engineers, buying servers, and renting space, but even after those expenses you still come out ahead if you have roughly more than 20 machines.

Except for the fact that hiring those engineers, buying those servers, and renting that space will be met with great resistance in most organizations. HR wont allow the extra headcount, legal is concerned about the safety of the space you're renting or some other BS, etc., etc. It's all a big headache and even if you decide to go through with it, it will take you weeks or even months to get off the ground.

But paying a monthly bill to Amazon? Nobody cares and you can have a new server running in minutes.

Re:Colocation?

[MattW]

(1) You just assumed that your use case was all use cases. Here's an example of a company I've done work for that is in the cloud. They have software that maps 2d face photos onto 3d models and then can render video out in flash that is customized for the user. (If you tried Nike's world cup you-in-the-advertisement video ad, that was them.) Can you imagine how variable demand is if your cpu utilization goes up 10,000% when a client pushes an ad to market, versus when you're largely idle and doing demos and handling light consumer demand? That's burstability.

(2) On a lesser scale, many sites may have huge variation on when they have high traffic. I'm sure we can find a lot of sites that are insanely busy in the evening from, say, 5-9pm, and basically idle all night long.

(3) Many organizations - larger ones - need to supply temporary environments, like dev & test/QA build environments. They need to spin them up to test applications and then dump them. They might be utilized 20 hours/month, but be under heavy load for those 20 hours. Did your calculations take into account needing an entire copy of your production site for 20 hours a day for dev & test?

This is just a few of many examples. Outside of Amazon, cloud computing also is used internally. I know of one large bank that got rid of 30,000 desktop machines in favor of 10,000 racked servers and KMS services. By day, the KMS services serve desktops to users, and IT never has to go touch an end user computer again. By night - so 15-16 hours a day - all that cpu, instead of sitting idle and being wasted, is scooped up by the software running their financial modeling software, so it number crunches all night long. It's like a triple win - easy to administer "desktops", less actual machines, and a bunch of extra compute power for their number crunching farm.

So, you may have done the math for your use case, but there are companies which could easily afford to build out datacenter space, and instead are using huge amounts of cloud resources instead.

Shut up

[tygerstripes]

Cloud Computing is such a loosely-defined and heavily abused term that its "true meaning" is almost as open to interpretation as "Web 2.0," and virtualised resources are often included in the definition.

The ever-colloquial Wikipedia states that it "typically involves over-the-Internet provision of dynamically scalable and often virtualized resources" while Foldoc states that it is "A loosely defined term for any system providing access via the Internet to processing power, storage, software or other computing services."

I'm fine with people debating the issue of the term's definition and provenance, even with people saying that one meaning is correct and another isn't, but flatly denying the existence of controversy without bothering to cite your authority is not conducive to anyone's understanding. Please, explain your position rather than simply stating it.

Come on George

[digitaldc]

Seriously, get some good writers BESIDES yourself and get actors who can actually ACT! And while you're at it, less CGI would be good and a couple more space battles....eh...what?
Oh, 'The Cloud Wars' isn't the title of the next Star Wars movie? Oh, sorry.

Cloud Wars

[dkleinsc]

It is a period of civil war. Rebel Linux admins, striking from a hidden base, have won their first victory against the evil Microsoft Empire. During the battle, GNU spies managed to steal secret plans to the Empire’s ultimate weapon, ISS, a system that brings any self-respecting admin to tears. Pursued by the Empire’s sinister agents, Tove Torvalds races home aboard her starship, custodian of the stolen plans that can save her people and restore freedom to the network.

Re:cloud vs VM

[sycorob]

As far as I can tell, what you're looking for is just virtual hosting, with a few specific requirements. I would think you could find all of those requirements, although I'll concede that a lot of options available today kind of suck.

"Cloud Computing" in my understanding is in fact all about the automatic scaling. I want to do a proof of concept online, and then show it to a few potential clients. I want it to basically be turned off when I'm not using it, and scale up quickly if my clients start hammering it. I only want to pay for what gets used. If I'm not using it, sell that capacity to somebody else, and keep my costs down.

If your're a full-fledged business, and you want responsiveness, and you want to guarantee a certain level of service to your clients, then cloud hosting may not be the best bet, or even the cheapest. You could still use it for special cases though. I heard a neat example: this guy needed to convert several million images into thumbnails. He wrote a little service to do it, hosted it in the cloud, let it scale way up, and churned through all the images in a few days, and it cost him a few hundred bucks on his corporate credit card. The time and expense to set up dedicated servers for this one-off task would have been ridiculous.

Like every damn thing in computer science (and really, life) cloud computing is not the solution to all of our problems, and it's also not a complete waste of time. It's a useful tool, to be used when appropriate.

Ooo! Car analogy! If you commute every day by car, you should probably buy a car. If you take the train to work and just need the car for the occasional trip to Target, then a car-sharing service might work for you.

Re:Huh?

[Dynedain]

Episodes 2&3 of what?

And why are you making a Yoda-like quote?

Can't solve for noisy neighbors with horizontal

[MattW]

You can't just keep scaling horizontally to avoid noisy neighbors. The problem is, unlike with cpu and memory, Amazon doesn't currently have a way to control how many IOPs one tenant has. You might even scale up from 2 "servers" to 4, and end up with the same neighbor because you're on the same underlying hardware. Plus, the issue is: it's not predictable. You might have great IOPs at one point, and then some other tenant starts consuming a bunch of them and there's contention, and your performance degrades.

No mention of Rack Space Cloud Sites

[phpsocialclub]

Rack Space cloud sites is true cloud hosting, You do not add servers, you just pay for CPU usage like you are buying bandwidth. It is expensive and you can not customize it, but it always works and scales to meet demand.

It is not a VM, where you can install your own OS, Web server, etc. It is a apache server, mysql clustered backend and varnish cache front end.

Great for Blog hosting, PHP applications, etc. We do about 8-10 million page views on it per month and like it for what it is,

If I had staff that needed to configure servers all day, it would be different, but for hosting large dynamic LAMP websites, it is great and reliable.

Andrew